MFCC for Dart 🎵

A pure Dart, null-safe library for extracting Mel-Frequency Cepstral Coefficients (MFCCs) from an audio signal. MFCCs are the most widely used features for speech recognition and audio analysis.

This package provides a flexible and easy-to-use API for both one-off calculations and real-time stream processing, with no platform-specific dependencies.

Features ✨

• Static Method: Quickly extract MFCCs from a complete signal.
• Instance-based Processing: Process audio frame-by-frame with a stateful processor.
• Stream Processing: Handle real-time audio streams seamlessly.
• Highly Customizable: Control all key parameters like sample rate, FFT size, number of filters, and coefficients.
• Standard Options: Includes pre-emphasis and log-energy replacement for the first coefficient.
• Pure Dart: Works on any platform where Dart runs (Mobile, Desktop, Web).
• Null-safe and well-documented.

---

Installation 💻

Add this to your project's pubspec.yaml file:

dependencies:
  mfcc: ^latest # Replace with the actual latest version

Then, run dart pub get or flutter pub get.

---

How to Use 🚀

There are three main ways to use this library, depending on your needs.

1. Static Method (Easiest)

This is the simplest way to get MFCCs from a complete audio signal.

import 'package:mfcc/mfcc.dart';
import 'dart:math';

void main() {
  // 1. Generate a sample audio signal (e.g., a 440 Hz sine wave)
  final int sampleRate = 16000;
  final signalDuration = 1; // seconds
  final signalLength = sampleRate * signalDuration;
  final signal = List<double>.generate(
      signalLength, (i) => sin(2 * pi * 440 * i / sampleRate));

  // 2. Define MFCC parameters
  final int windowLength = 400; // ~25 ms
  final int windowStride = 160; // ~10 ms
  final int fftSize = 512;
  final int numFilters = 40;
  final int numCoefs = 13;

  // 3. Extract features
  List<List<double>> features = MFCC.mfccFeats(
    signal,
    sampleRate,
    windowLength,
    windowStride,
    fftSize,
    numFilters,
    numCoefs,
    energy: true,
    preEmphasis: 0.97,
  );

  // 4. Print the results (num_frames x num_coeffs)
  print('Extracted ${features.length} frames.');
  print('Each frame has ${features[0].length} coefficients.');
  // Example: print the first frame's features
  print('First frame features: ${features[0]}');
}

2. Instance-based Processing

This approach is ideal when you need to process audio frame by frame.

import 'package:mfcc/mfcc.dart';

void main() {
  // 1. Instantiate and configure the MFCC processor
  final processor = MFCC(
    sampleRate: 16000,
    fftSize: 512,
    numFilters: 40,
    numCoefs: 13,
  );

  // 2. Assume you have pre-framed audio data
  // (e.g., from a microphone buffer)
  final List<List<double>> audioFrames = [
    List<double>.filled(400, 0.5), // Frame 1
    List<double>.filled(400, 0.3), // Frame 2
    List<double>.filled(400, 0.8), // Frame 3
  ];

  // 3. Process all frames at once
  List<List<double>> features = processor.processFrames(audioFrames);
  print('Processed ${features.length} frames.');
  print('Features for frame 1: ${features[0]}');

  // Or process frames one by one
  List<double> singleFrameFeatures = processor.processFrame(audioFrames[0]);
  print('Features for frame 1 (processed individually): $singleFrameFeatures');
}

3. Stream Processing

Use streams for real-time applications, like processing audio from a microphone.

import 'package:mfcc/mfcc.dart';
import 'dart:async';

void main() async {
  // 1. Instantiate the MFCC processor
  final processor = MFCC(
    sampleRate: 16000,
    fftSize: 512,
    numFilters: 40,
    numCoefs: 13,
  );

  // 2. Create a mock stream of audio frames (in a real app, this would
  // come from a source like `mic_stream`)
  final audioFrameStream = Stream.periodic(
    const Duration(milliseconds: 100),
    (i) => List<double>.filled(400, i * 0.1),
  ).take(5);

  // 3. Set the stream and get the output feature stream
  final featureStream = processor.setStream(audioFrameStream);

  // 4. Listen for new MFCC features as they are computed
  if (featureStream != null) {
    print('Listening for MFCC features...');
    await for (final features in featureStream) {
      print('New features computed: [${features.map((e) => e.toStringAsFixed(2)).join(', ')}]');
    }
  }
  
  // 5. Clean up when done
  processor.cancelStream();
  print('Stream processing finished.');
}

---

API Overview 📖

• MFCC class: The main class for processing.
  • Constructor: MFCC({ required int sampleRate, ... }) configures the processor.
  • processFrame(List<double> frame): Processes a single frame.
  • processFrames(List<List<double>> frames): Processes a list of frames.
  • setStream(Stream<List<double>> audioInput): Sets up real-time stream processing.
  • cancelStream(): Stops and cleans up stream resources.
• MFCC.mfccFeats(...): Static helper for a one-off calculation on a full signal.
• ValueError: Custom exception for invalid parameters.

---

License

This project is licensed under the GNU Affero General Public License v3.0. See the LICENSE file for details.

---

Contributing

Contributions are welcome! If you find a bug or want to suggest a new feature, please open an issue.